adding tests for a variety of sparse dtypes

theano/sparse/basic.py

@@ -696,7 +696,7 @@ class StructuredDot(gof.Op):
             else:
                 raise Exception("a.shape=%s, b.shape=%s, variable.shape=%s ??? I have no idea why")
 
-        ## Commenting this out because variable should be a numpy.ndarray since the assert above
+        ## Commenting this out because variable should be a numpy.ndarray since the "assert _is_dense(variable)" above
         ## (JB 20090109)
         # out[0] = numpy.asarray(variable)  #TODO: fix this really bad implementation
         #
@@ -714,6 +714,7 @@ def structured_dot(x, y):
     """
     @todo: Maybe the triple-transposition formulation (when x is dense)
     is slow. See if there is a direct way to do this.
+    (JB 20090528: Transposing tensors and sparse matrices is constant-time, inplace, and fast.)
     """
     if hasattr(x, 'getnnz'): x = as_sparse_variable(x)
     if hasattr(y, 'getnnz'): y = as_sparse_variable(y)

theano/sparse/tests/test_basic.py

@@ -161,18 +161,25 @@ class test_structureddot(unittest.TestCase):
 
     def test_structuredot(self):
         bsize = 2
+        typenames = 'int8', 'int16', 'int32', 'int64', 'float32', 'float64', 'complex64', 'complex128'
        
-        # iterate 10 times just to make sure (cannot get this wrong !)
+        for sparse_dtype in typenames:
+           for dense_dtype in typenames:
+                # iterate for a few different random graph patterns
                 for i in range(10):
-            spmat = sp.lil_matrix((4,6))
+                    spmat = sp.lil_matrix((4,6), dtype=sparse_dtype)
                     for i in range(5):
+                        # set non-zeros in random locations (row x, col y)
                         x = numpy.floor(numpy.random.rand()*spmat.shape[0])
                         y = numpy.floor(numpy.random.rand()*spmat.shape[1])
                         spmat[x,y] = numpy.random.rand()*10
                     spmat = sp.csc_matrix(spmat)
                
-            kerns = tensor.dvector('kerns')
-            images = tensor.dmatrix('images')
+                    kerns = tensor.Tensor(broadcastable=[False], dtype=sparse_dtype)('kerns')
+                    images = tensor.Tensor(broadcastable=[False, False], dtype=dense_dtype)('images')
+
+                    output_dtype = theano.scalar.upcast(sparse_dtype, dense_dtype)
+                    assert output_dtype in (sparse_dtype, dense_dtype)
 
                     ##
                     # Test compressed-sparse column matrices ###
@@ -181,6 +188,7 @@ class test_structureddot(unittest.TestCase):
                     # build symbolic theano graph
                     def buildgraphCSC(kerns,images):
                         csc = CSC(kerns, spmat.indices[:spmat.size], spmat.indptr, spmat.shape)
+                        assert csc.type.dtype == output_dtype
                         return structured_dot(csc, images.T)
                     out = buildgraphCSC(kerns,images)
                     f = theano.function([kerns,images], out)
@@ -192,7 +200,10 @@ class test_structureddot(unittest.TestCase):
                     c = spmat * (imvals.T)
                     assert _is_dense(c)
                     assert numpy.all(outvals == c)
+                    assert str(outvals.dtype) == output_dtype
+                    assert c.dtype == outvals.dtype
 
+                    if sparse_dtype.startswith('float') and dense_dtype.startswith('float'):
                         utt.verify_grad(buildgraphCSC, [kernvals,imvals])
 
                     ##
@@ -214,7 +225,11 @@ class test_structureddot(unittest.TestCase):
                     c = spmat * (imvals.T)
                     assert _is_dense(c)
                     assert numpy.all(outvals == c)
+                    assert str(outvals.dtype) == output_dtype
+                    assert c.dtype == outvals.dtype
 
+                    # we could test more, but hopefully this suffices?
+                    if sparse_dtype.startswith('float') and dense_dtype.startswith('float'):
                         utt.verify_grad( buildgraphCSR, [kernvals,imvals])